Back to skill

Security audit

Daxiang Agent Dispatch

Security checks for vulnerabilities and agentic risk

Overview

This skill is a disclosed agent router, but it needs review because it can automatically send broad user requests and raw prompt content to other agents while logging routing metadata.

Install only if you want a central dispatcher that can automatically route your requests to multiple specialist agents. Avoid sending secrets or sensitive personal/business data unless you are comfortable with that content being passed to other agents, and review or tighten the routing patterns and logging policy before use in a sensitive environment.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (9)

Self-Modification

High
Category
Rogue Agent
Confidence
85% confidence
Finding

Skill modifies its own code, configuration, or behavior at runtime. Self-modification enables an agent to escalate privileges, disable safety constraints, or install persistent backdoors.

Content

Scanner excerpt · SKILL.md (reported line 1)May include surrounding context.

md
# Agent Dispatch Skill

**版本**: v1.0
**创建日期**: 2026-03-26

Vague Triggers

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The catch-all pattern '.*' matches every request, making routing fully permissive and guaranteeing that any unmatched or malformed input is accepted and forwarded. In a multi-agent system, this increases the risk of prompt injection propagation, unintended capability exposure, and suppression of safe failure behavior because no request is ever rejected for being out of scope.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The skill forwards the user's original input to spawned specialist agents, but it does not disclose that user content may be shared across agent boundaries. In a multi-agent system, this can expose sensitive prompts, secrets, personal data, or proprietary content to additional execution contexts without informed consent or data-minimization controls.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documentation at L057-L061 says routing stops at the first matching rule after checking in priority order. However, in the illustrated routing list the catch-all pattern .* appears at L050-L054 before later category rules are fully shown, and the implementation at L112-L119 performs a simple first-match scan. This means the documented intent of priority-based semantic dispatch can be contradicted by the actual listed rule order.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The logging design records user activity metadata, routing choices, outcomes, and timing information for monitoring without any notice, retention policy, or privacy guardrails. Even if raw user_input is not shown in the sample log entry, this telemetry can still reveal sensitive behavior patterns and may be combined with other logs to reconstruct user actions.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The manifest uses generic trigger terms like "search," "research," "study," and "learn" that commonly appear in ordinary conversation, making routing to this skill potentially over-broad. The configuration does not provide narrowing context, exclusion conditions, or negative examples to clarify when this route should or should not activate.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger pattern includes very common words such as "write" and "content," which overlap heavily with normal user requests and do not clearly distinguish when this routing rule should apply. Because the manifest lacks additional constraints or negative examples, the activation boundary is ambiguous.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

Terms such as "task" and "manage" are generic and likely to appear in many unrelated requests, which can cause unintended routing. The manifest does not define scope limits or exclusion conditions that would make activation behavior predictable.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

Nearly all user-facing instructional content in the skill is presented only in Chinese, and there is no indication that users may choose another language. Under the language/locale policy, a skill should not impose a specific language without opt-in unless the locale restriction is explicitly justified.

Content

No source excerpt is available for this finding.